Fx

silero-vad.ipynb

@@ -301,7 +301,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.8.3"
+   "version": "3.7.7"
   },
   "toc": {
    "base_numbering": 1,

utils.py

@@ -73,11 +73,15 @@ def init_jit_model(model_path,
     return model
 
 
-def get_speech_ts(wav, model, extractor, trig_sum=0.25, neg_trig_sum=0.01, num_steps=8, batch_size=200):
+def get_speech_ts(wav, model, extractor,
+                  trig_sum=0.25, neg_trig_sum=0.01,
+                  num_steps=8, batch_size=200):
+
     assert 4000 % num_steps == 0
-    step = int(4000 / num_steps)
+    step = int(4000 / num_steps)  # stride / hop
     outs = []
     to_concat = []
+
     for i in range(0, len(wav), step):
         chunk = wav[i: i+4000]
         if len(chunk) < 4000:
@@ -98,11 +102,12 @@ def get_speech_ts(wav, model, extractor, trig_sum=0.25, neg_trig_sum=0.01, num_s
 
     outs = torch.cat(outs, dim=0)
 
-    buffer = deque(maxlen=num_steps)
+    buffer = deque(maxlen=num_steps)  # when max queue len is reach, first element is dropped
     triggered = False
     speeches = []
     current_speech = {}
-    for i, predict in enumerate(outs[:, 1]):
+
+    for i, predict in enumerate(outs[:, 1]):  # add name
         buffer.append(predict)
         if (np.mean(buffer) >= trig_sum) and not triggered:
             triggered = True
@@ -150,7 +155,9 @@ class STFTExtractor(nn.Module):
 
 
 class VADiterator:
-    def __init__(self, trig_sum=0.26, neg_trig_sum=0.01, num_steps=8):
+    def __init__(self,
+                 trig_sum=0.26, neg_trig_sum=0.01,
+                 num_steps=8):
         self.num_steps = num_steps
         assert 4000 % num_steps == 0
         self.step = int(4000 / num_steps)
@@ -203,7 +210,6 @@ class VADiterator:
         return current_speech, self.current_name
 
 
-
 def state_generator(model, audios, extractor, onnx=False, trig_sum=0.26, neg_trig_sum=0.01, num_steps=8, audios_in_stream=5):
     VADiters = [VADiterator(trig_sum, neg_trig_sum, num_steps) for i in range(audios_in_stream)]
     for i, current_pieces in enumerate(stream_imitator(audios, audios_in_stream)):
@@ -259,4 +265,3 @@ def stream_imitator(stereo, audios_in_stream):
             values.append((out, wav_name))
         yield values
 
-